Click here to

Session: Therapy: Outcome Modeling and Assessment II [Return to Session]

A Functional Dosiomics-Based Prediction Model for Radiation Pneumonitis

B Liang*, Y Tian, R Wei, J Miao, P Huang, Z Liu, W Xia, J Dai, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, BeijingCN,


MO-IePD-TRACK 5-1 (Monday, 7/26/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]

Purpose: Ventilation image (VI) reveals the presence of functional heterogeneity within lungs. This study adopted the omics method to mine the heterogeneous information to better predict the incidence of radiation pneumonitis (RP).

Methods: VI was derived from the 4-dimensional computed tomography (4DCT). The functional dose (FD) distribution was obtained by weighting OD with VI. The omics features were extracted, and then selected by the minimum redundancy maximum relevance (mRMR) method. The prediction models were constructed with the logistic regression (LR) , support vector machine (SVM), k-nearest neighbors (KNN), and random forest (RF) classifiers. Cross validation, bootstrap and nested sampling and methods were adopted for model training and classifier parameter hyper-tuning. The prediction models based on the omics features extracted from OD and he dosimetric factors (V5, V20 and mean lung dose, DF) based model were constructed for comparison.

Results: 217 thoracic cancer patients treated with radiotherapy were used to train and validate the prediction model. The 4DCT-based VI showed the heterogeneous pulmonary function of the lungs. 19 first-order, 27 gray level (GL) co-occurrence matrix (GLCM), 16 GL run length matrix (GLRLM) and 16 GL size zone (GLSZM) features were extracted from the ipsilateral, contralateral and total lungs, separately. In total, 234 (78×3) features were extracted for each patient. The FD-omics and OD-omics models achieved optimal performance when 5 features were selected using the LR classifier. And the area under curve (AUC) were 0.755 (0.753-0.757), 0.716 (0.713-0.718), respectively. In comparison, the AUC of the DF model was 0.680 (0.675-0.684).

Conclusion: The FD-omics model outperformed the OD-omics and DF modes for RP prediction due to the use of more data points and the data-mining in greater depth. It also indicates that the heterogeneity of pulmonary distribution revealed by VI helps to better model the dose response of radiation.

Funding Support, Disclosures, and Conflict of Interest: The work was supported by the National Natural Science Foundation of China (11875320, 81801799 and 81502649).





    TH- Response Assessment: Radiomics/texture/feature-based response assessment

    Contact Email